A text semantic computing and cognitive input evaluation method based on conceptual representation

Abstract

The application discloses a text semantic calculation and cognitive input evaluation method based on conceptual representation. Starting from the perspective of psycholinguistics and based on the language features of educational concepts, the application formally describes two important dimensions of the familiarity and abstractness of educational concepts, quantitatively calculates the attribute values of the familiarity and abstractness of educational concepts, and constructs a two-dimensional attribute model to better understand and apply educational concepts. The constructed "familiarity-abstractness (F-A)" model for educational concepts effectively maps educational concepts into a two-dimensional space. By quantitatively calculating the psychological and linguistic attribute values of educational concepts, online discussion text conceptualization is further realized. With the help of the model, the text semantic complexity of online discussion text can be accurately calculated, so that the cognitive input state and cognitive processing process of learners can be more deeply understood, and the cognitive input level of online discussion text can be further automatically evaluated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of text information processing, and in particular to a text semantic calculation and cognitive input evaluation method based on conceptual representation. BACKGROUND

[0002] In a large-scale online learning platform, the text semantic complexity of posts published by learners is a predictor of the cognitive input level of learners. When learners publish posts with complex content involving multiple viewpoints and domain concepts in discussions, it indicates that they have conducted in-depth thinking and exploration, that is, their cognitive input level is high. On the contrary, if the content of the posts published by the learners is relatively simple and lacks depth, it indicates that the cognitive input of the learners is low. Thus, the text semantic complexity of online discussion posts can be used as an indicator to measure the cognitive input of learners.

[0003] In the field of education, understanding and representing the cognitive characteristics of educational concepts are crucial for effective teaching and learning. Educational concepts are the basic units in the field of education and an important part of the knowledge system of disciplines. Understanding and representing the cognitive characteristics of educational concepts help teachers and learners to deeply understand the nature of disciplines and grasp the structure of knowledge, thereby improving the effectiveness of teaching and learning.

[0004] Current research in the field of educational concept representation relies on general corpora, and text complexity analysis is mostly coarse-grained, so there is a lack of in-depth analysis of educational concepts, and the semantic information of educational concepts has not been fully mined, resulting in weak pertinence and inability to be interpreted. SUMMARY

[0005] Therefore, an embodiment of the present application provides a text semantic calculation and cognitive input evaluation method based on conceptual representation.

[0006] The first aspect of the present application provides a text semantic calculation and cognitive input evaluation method based on conceptual representation, comprising the following steps:

[0007] An educational concept dictionary is constructed, and the educational concept dictionary is used to match educational concepts in online discussion texts to obtain candidate educational concepts of online discussion texts;

[0008] The BERT model is used to calculate the semantic relevance of online discussion texts and the candidate educational concepts, and the candidate educational concepts with a semantic relevance lower than a preset relevance threshold are filtered;

[0009] An educational concept F-A model is constructed according to the educational concept dictionary, and the complexity and abstractness of the candidate educational concepts are obtained from the educational concept F-A model;

[0010] According to the complexity and abstraction of the candidate educational concept, the semantic complexity of the online discussion text is calculated;

[0011] The cognitive input level and the semantic complexity of the online discussion text are analyzed, and a model is trained to automatically identify the cognitive input level of the online discussion text.

[0012] Further, the educational concept dictionary is constructed, specifically including the following steps:

[0013] A plurality of educational concepts are extracted from a preset corpus;

[0014] The educational concepts are stored and organized using a Trie tree data structure to form an educational concept dictionary.

[0015] Further, the educational concept dictionary is used to match the educational concepts in the online discussion text to obtain candidate educational concepts of the online discussion text, specifically including the following steps:

[0016] The online discussion text is processed by word segmentation;

[0017] The edit distance between the segmented online discussion text and each educational concept in the educational concept dictionary is calculated, and the educational concept with an edit distance lower than a preset distance threshold is taken as a candidate educational concept of the online discussion text.

[0018] Further, the BERT model is used to calculate the semantic relevance of the online discussion text and the candidate educational concept, specifically including the following steps:

[0019] The BERT model is used to construct a vector space of the online discussion text and the candidate educational concept;

[0020] The semantic relevance of each candidate educational concept and the online discussion text is calculated using the following formula:

[0021]

[0022] In the formula, V P represents the vector space of the online discussion text, V Ci represents the vector space of each candidate educational concept, i∈[1,j], and j is the total number of candidate educational concepts.

[0023] Further, the candidate educational concept with a semantic relevance lower than a preset relevance threshold is filtered, specifically including the following steps:

[0024] A candidate educational concept set is constructed according to the candidate educational concept;

[0025] The average value and the standard deviation of the candidate educational concept set are determined to determine the preset relevance threshold.

[0026] Further, the educational concept F-A model is constructed according to the educational concept dictionary, and the complexity and the abstract degree of the candidate educational concept are obtained from the educational concept F-A model, and specifically comprising the following steps:

[0027] The educational concept in the educational concept dictionary is represented by a feature vector, and a language feature vector of the educational concept is constructed;

[0028] Regression analysis is performed according to the language feature vectors of a plurality of educational concepts, the familiarity and the abstract degree of each educational concept are determined, and an educational concept F-A model is established;

[0029] The complexity and the abstract degree of the candidate educational concept are obtained in the educational concept F-A model.

[0030] Further, the semantic complexity of the online discussion text is calculated according to the complexity and the abstract degree of the candidate educational concept, and specifically comprising the following steps:

[0031] The online discussion text is represented as a set of candidate educational concepts;

[0032] The semantic complexity of the online discussion text is calculated using the following formula:

[0033]

[0034] Wherein, N represents the total number of candidate educational concepts, i represents each candidate educational concept, i∈[1,N]; W F represents the familiarity weight of the candidate educational concept, F(C i ) represents the familiarity of the candidate educational concept in the educational concept F-A model; W A represents the abstract degree weight of the candidate educational concept, A(C i ) represents the familiarity of the candidate educational concept in the educational concept F-A model.

[0035] Further, the familiarity weight and the abstract degree weight of the candidate educational concept are calculated by the quasi-Newton method.

[0036] Further, the cognitive input level and the semantic complexity of the online discussion text are analyzed, and specifically comprising the following steps:

[0037] The cognitive input level of the online discussion text is encoded;

[0038] The online discussion text is described using the correlation analysis tool, and the correlation between the online discussion text and the semantic complexity and the cognitive input level is obtained.

[0039] Further, the training model automatically identifies the cognitive input level of the online discussion text, and specifically includes the following steps of determining the target machine learning model of training.

[0040] The semantic complexity is used as an input feature vector, and the cognitive input level is used as a model output to construct a training data set.

[0041] The model is trained using the training data set to obtain a machine learning model capable of automatically identifying the cognitive input level of the online discussion text.

[0042] The embodiment of the application also discloses a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. The processor of the computer device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions to enable the computer device to execute the foregoing method.

[0043] The embodiment of the application has the following beneficial effects: the text semantic calculation and cognitive input evaluation method based on conceptual representation formalizes the two important dimensions of familiarity and abstractness of the educational concept based on the language features of the educational concept from the perspective of psycholinguistics, quantitatively calculates the attribute values of the familiarity and abstractness of the educational concept, and constructs a two-dimensional attribute model to better understand and apply the educational concept. The constructed familiarity-abstractness (F-A) model for the educational concept effectively maps the educational concept to a two-dimensional space. By quantitatively calculating the psychological language attribute values of the educational concept, the conceptualization of the online discussion text is further realized. With the help of the model, the text semantic complexity of the online discussion text can be accurately calculated, so that the cognitive input state and cognitive processing process of the learner can be better understood, and the cognitive input level of the online discussion text can be further automatically evaluated.

[0044] Additional aspects and advantages of the application will be described in the following description section, some of which will become apparent from the following description, or will be understood through practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0046] Figure 1 It is a basic implementation flowchart of the text semantic calculation and cognitive input evaluation method based on conceptual representation.

[0047] Figure 2 is an effect diagram of an educational concept dictionary for constructing a Trie tree data structure according to the present application.

[0048] Figure 3 is a structure diagram of a BERT model used in the present application.

[0049] Figure 4 is an effect diagram of an F-A model constructed in the present application.

[0050] Figure 5 is an attribute control diagram of an F-A model constructed in the present application.

[0051] Figure 6 is an error bar chart of cognitive input level and semantic complexity according to the present application.

[0052] Figure 7 is an effect diagram of training a machine learning model using a training data set according to the present application. DETAILED DESCRIPTION

[0053] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0054] In order to more objectively and accurately define and describe the discussion text of learners, the embodiment of the present application provides a text semantic calculation and cognitive input evaluation method based on conceptual representation. As shown in Figure 1 the embodiment of the present application, the method comprises the following steps:

[0055] S1. Construct an educational concept dictionary, match the educational concepts in the online discussion text using the educational concept dictionary, and obtain the candidate educational concepts of the online discussion text.

[0056] In step S1, the embodiment of the present application constructs the educational concept dictionary as a Trie tree structure, matches the online discussion text after tokenization with the educational concept dictionary Trie tree, realizes the deduction of the educational concepts related to the online discussion text from the online discussion text, and forms a candidate concept set.

[0057] S2. Calculate the semantic relevance of the online discussion text and the candidate educational concepts using the BERT model, and filter the candidate educational concepts whose semantic relevance is lower than the preset relevance threshold.

[0058] In step S2, the embodiment of the present application utilizes the BERT pre-training model to construct a vector space of the online discussion text and all candidate educational concepts, and calculates the semantic correlation of each candidate educational concept with the online discussion text, to obtain a semantic correlation list of the online discussion text and the educational concepts. In addition, the threshold of the correlation list is determined by using the central limit theorem, and the candidate educational concepts with a correlation lower than the threshold are filtered out.

[0059] S3. Constructing an educational concept F-A model according to the educational concept dictionary, and obtaining the complexity and abstractness of the candidate educational concept from the educational concept F-A model.

[0060] In step S3, the embodiment of the present application constructs the educational concept F-A model, to realize the vector space formalization representation of the psychological and linguistic attributes of the educational concept. The educational concept is projected into a two-dimensional space through the F-A model, so as to depict the psychological and linguistic attributes of the educational concept, to better understand and diagnose the cognitive input state and cognitive processing process of the learner in understanding the concept.

[0061] S4. Calculating the semantic complexity of the online discussion text according to the complexity and abstractness of the candidate educational concept.

[0062] In step S4, the embodiment of the present application utilizes the familiarity and abstractness of the educational concept contained in the online discussion text, to calculate the semantic complexity of the online discussion text.

[0063] S5. Correlation analysis of the cognitive input level and the semantic complexity of the online discussion text; training a model to automatically identify the cognitive input level of the online discussion text.

[0064] In step S5, the embodiment of the present application obtains the cognitive input level of the online discussion text through artificial coding based on the ICAP framework of cognitive input, with a total of 4 levels from low to high. The cognitive input of the text is correlated with the semantic complexity, to explore the semantic complexity characteristics of the online discussion text with different cognitive inputs, and to select appropriate feature vectors to train a model for evaluating the cognitive input level of the online discussion text.

[0065] The method of the present application effectively maps the educational concept into a two-dimensional space by constructing a familiarity-abstractness (F-A) model for the educational concept. By quantitatively calculating the psychological and linguistic attribute values of the educational concept, the conceptualization of the online discussion text is further realized, so that the discussion text of the learner can be more objectively and accurately defined and described. With the help of this model, the text semantic complexity of the online discussion text can be accurately calculated, so as to better understand the cognitive input state and cognitive processing process of the learner, and the cognitive input level of the online discussion text can be further automatically evaluated.

[0066] As a preferred embodiment, the implementation process of each step of the method of the present application is discussed in detail below:

[0067] S1. Construct an education concept dictionary, and use the education concept dictionary to match the education concepts in the online discussion text to obtain candidate education concepts of the online discussion text.

[0068] In step S1, the education concept dictionary is constructed, specifically including the following steps:

[0069] S1-1. Extract a plurality of education concepts from a preset corpus.

[0070] In the embodiment of the present application, the corpus is crawled from MOOCCube (a large corpus covering different fields), MOOC (Massive Open Online Course), an education literature library (such as Baidu Knows, major libraries, etc.), and the like, to obtain the online discussion text of the learner. The online discussion text is subjected to education concept extraction to form a dictionary containing m education concepts, which is defined as D={C1,C2,C3,…,Cm}. m-1 ,C m}.

[0071] S1-2. Use a Trie tree data structure to store and organize the education concepts to form the education concept dictionary.

[0072] The Trie tree data structure is shown in Figure 2 . The dictionary Trie tree starts from a root node, which is used as the starting point of all words; the first word of the education concept is inserted into the root node; then the subsequent word of the education concept is added one by one to the dictionary tree until the last character of the education concept is reached. If the corresponding word already exists in the dictionary tree, the subnode is transferred and the next word is processed. When the end of the education concept is reached, a special key (EOT) is inserted to identify the end of the term.

[0073] In step S1, the education concept dictionary is used to match the education concepts in the online discussion text to obtain candidate education concepts of the online discussion text, specifically including the following steps:

[0074] S1-3. Perform word segmentation processing on the online discussion text.

[0075] In the embodiment of the present application, the existing natural language processing tools such as jieba, ltp, IKAnalyzer, and the like are used to perform word segmentation on the crawled online discussion text P.

[0076] S1-4. Calculate the edit distance between the segmented online discussion text and each education concept in the education concept dictionary, and take the education concept with an edit distance lower than a preset distance threshold as the candidate education concept of the online discussion text.

[0077] In this embodiment of the invention, edit distance (Levenshtein distance) is used to generate a candidate concept set.

[0078] Edit distance refers to the minimum number of edit operations required to transform one string into another, and can measure the orthogonal similarity between two target objects. It measures the educational concepts in the Trie tree of the segmented online discussion text P and the educational concepts in the trie tree. The maximum edit distance is set to 1, and the relevant educational concepts C in the online discussion text P are derived, generating a candidate concept set for the online discussion text P, i.e., P = {C1, C2, C3, ..., C}. j-1 C j Let j be the number of candidate concepts. Examples are shown in Table 1.

[0079] Table 1. Examples of concept matching in online discussion texts

[0080]

[0081] S2. Use the BERT model to calculate the semantic relevance between online discussion text and candidate educational concepts, and filter candidate educational concepts whose semantic relevance is lower than the preset relevance threshold.

[0082] In step S2, the semantic relevance between the online discussion text and the candidate educational concepts is calculated using the BERT model, specifically including the following steps:

[0083] S2-1. Construct a vector space for online discussion texts and candidate educational concepts using the BERT model;

[0084] like Figure 3 As shown, BERT stands for Bidirectional Encoder Representations from Transformer. The goal of BERT is to obtain semantic representations of text through training on large-scale unlabeled corpora, then fine-tune these semantic representations for specific NLP (Natural Language Processing) tasks, and finally apply them to those NLP tasks. In deep neural network-based NLP methods, characters / words in text are typically represented by one-dimensional vectors. The neural network takes the one-dimensional word vectors of each character or word as input, performs a series of complex transformations, and outputs a one-dimensional word vector as the semantic representation of the text. Specifically, we typically want semantically similar characters / words to be close in distance in the feature vector space, so that the text vectors derived from character / word vectors can contain more accurate semantic information.

[0085] This invention employs a BERT pre-trained model to measure the semantic relevance between candidate educational concepts and online discussion text, filtering out candidate educational concepts with relevance below a threshold. For example, given an online discussion text P, its corresponding set of candidate educational concepts is {C1, C2, C3, ..., C...}. j-1 C j Using a BERT pre-trained model, a vector space is constructed for post P and all candidate educational concepts, yielding the corresponding vector V. P and

[0086] In some embodiments, in addition to the BERT model, other derived models, such as word2vec, Tencent AI Lab, and ERNIE models, can be used for semantic vector representation extraction.

[0087] S2-2. Calculate the semantic relevance of each candidate educational concept to the online discussion text using the following formula:

[0088]

[0089] In the formula, V P V represents the vector space of online discussion text. Ci Let i represent the vector space of each candidate educational concept, i∈[1,j], where j is the total number of candidate educational concepts.

[0090] Using the semantic relevance formula, we can calculate the following list R = {R1, R2, R3, ..., R...}, consisting of j semantic relevance values. j-1 ,R j};

[0091]

[0092] In step S2, candidate educational concepts with semantic relevance below a preset relevance threshold are filtered, specifically including the following steps:

[0093] S2-3. Construct a set of candidate educational concepts based on the candidate educational concepts.

[0094] S2-4. Determine the mean and standard deviation of the candidate educational concept set to determine the preset relevance threshold.

[0095] In this embodiment of the invention, based on the average value of the relevance list R... The lower threshold is determined using the standard deviation σ, and candidate educational concepts with relevance below the threshold are filtered out. That is, each post can consist of N educational concepts with high semantic relevance to the post, P = {C1, C2, C3, ..., C...}. N-1 C N}

[0096] S3. Constructing an educational concept F-A model according to the educational concept dictionary, and obtaining the complexity and abstractness of the candidate educational concept from the educational concept F-A model.

[0097] In step S3, the educational concept F-A model is constructed according to the educational concept dictionary, and the complexity and abstractness of the candidate educational concept are obtained from the educational concept F-A model. Specifically, the following steps are included:

[0098] S3-1. The educational concepts in the educational concept dictionary are represented by feature vectors, and the language feature vectors of the educational concepts are constructed.

[0099] In the embodiment of the present application, the educational concepts in the educational concept dictionary D={C1, C2, C3, …, Cn} are represented by feature vectors in terms of word length, pinyin, strokes, part of speech, and semantics, i.e. m-1 m Many studies have shown that the language features of Chinese words such as word length, pinyin, strokes, and semantics are closely related to their psychological language properties. By representing the educational concepts by feature vectors, the psychological language properties of the educational concepts can be better inferred.

[0100] S3-2. Regression analysis is performed according to the language feature vectors of the multiple educational concepts to determine the familiarity and abstractness of each educational concept, and an educational concept F-A model is established.

[0101] In the embodiment of the present application, the psychological language properties of the educational concept C are described as a two-dimensional attribute value vector L(C)=(F, A), where F and A represent the values of familiarity and abstractness, respectively, and the value range is between 1 and 7 (when the attribute value is 1, it means very unfamiliar or very specific; when the attribute value is 7, it means very familiar or very abstract).

[0102] ​​Familiarity (F) and Abstractness (A) are both psychological linguistic attributes of text components such as vocabulary, concept, sentence, etc., and are important factors affecting the cognitive processing and modeling calculation of learners on vocabulary, domain concept, sentence, and discourse. Familiarity refers to the familiarity of learners with the concept, including the understanding of the definition, instances, attributes, and relevance of the concept. Familiarity can be measured by the prior knowledge, experience, and familiarity of learners with the concept. High familiarity means that learners have more prior knowledge and experience of the concept, while low familiarity means that learners are relatively unfamiliar with the concept or lack sufficient prior knowledge. Abstractness refers to the depth and breadth of thinking involved in the concept, which can be approached from the connotation and extension of the concept. Connotation refers to the essential attributes and core features of the concept, while extension refers to the specific instances and situations covered by the concept. Concepts with high abstractness usually have fewer real instances and situations, while concepts with low abstractness are the opposite.

[0103] The language feature vector of the educational concept C in the embodiment of the application The familiarity and abstractness psychological linguistic attribute model L(C) of the educational concept is calculated as (F, A), and the F-A model is as shown in Figure 4

[0104] Specifically, the embodiment of the application first annotates the familiarity F and abstractness A of a part of educational concepts, and then maps the educational concept language feature vector FV to the predicted values of the psychological linguistic attributes of familiarity F and abstractness A through a regression analysis model (such as a Catboost, LGBT, etc. integrated learning model), to determine the familiarity F value and abstractness A value of all educational concepts, and finally obtain L(C) = (F, A).

[0105] S3-3. Obtain the complexity and abstractness of the candidate educational concept in the educational concept F-A model.

[0106] In the embodiment of the application, the familiarity F and abstractness A attribute values of the educational concept are divided into two parts from low to high, forming a "two-dimensional four-quadrant" psychological linguistic attribute space. Each educational concept C can find the corresponding coordinates in this space, as shown in Figure 5 Figure 5 The tree control shown provides a basis for in-depth analysis and calculation of the psychological linguistic attribute values of educational texts (such as online discussion texts, student essays, teacher comments, etc.). The psychological linguistic attributes of educational concepts can better understand and diagnose the cognitive input state and cognitive processing of learners on the concept.

[0107] S4. Calculate the semantic complexity of the online discussion text according to the complexity and abstractness of the candidate educational concept. ​​

[0108] In step S4, the semantic complexity of the online discussion text is calculated according to the complexity and abstractness of the candidate educational concepts, and specifically includes the following steps:

[0109] S4-1. Represent the online discussion text as a set of candidate educational concepts.

[0110] In the embodiment of the present application, the online discussion text P is represented as a set of N high-relevance educational concepts, i.e. P = {C1, C2, C3, …, C N-1 , N} after the word segmentation processing, the matching of educational concepts and the filtering of relevance detection.

[0111] At the same time, since the familiarity and abstractness of the domain terms are important indicators affecting the semantic complexity of the text, the familiarity and abstractness of the educational concepts contained in the online discussion text can be used to estimate the semantic complexity of the text.

[0112] S4-2. Calculate the semantic complexity of the online discussion text using the following formula:

[0113]

[0114] wherein N represents the total number of candidate educational concepts, i represents each candidate educational concept, i ∈ [1, N]; W F represents the familiarity weight of the candidate educational concept, F(C i ) represents the familiarity of the candidate educational concept in the F-A model of educational concepts; W A represents the abstractness weight of the candidate educational concept, A(C i ) represents the abstractness of the candidate educational concept in the F-A model of educational concepts.

[0115] In the embodiment of the present application, the corresponding values of F(C i ) and A(C i ) can be found in the L(C) = (F, A) of step S3 through the corresponding psychological language attribute values of the concept C i in the educational concept dictionary. The N, W F , W A , F(C i ), and A(C i ) are input into the formula , and the semantic complexity result of the online discussion text can be output.

[0116] As a preferred embodiment, the familiarity weight W F and the abstractness weight W A of the candidate educational concept in the semantic complexity formula are calculated by the Quasi-Newton Methods.

[0117] S5. Correlation analysis is performed on the cognitive input level and semantic complexity of the online discussion text; and a model is trained to automatically identify the cognitive input level of the online discussion text.

[0118] In step S5, the cognitive input level and semantic complexity of the online discussion text are correlated and analyzed, specifically including the following steps:

[0119] S5-1. The cognitive input level of the online discussion text is coded.

[0120] In the embodiment of the present application, the cognitive input level of the online discussion text is coded, mainly based on the ICAP cognitive input framework. The ICAP framework is a theoretical model for understanding and analyzing the cognitive input level of learners in the learning process. ICAP is an acronym of four English words, which represent the following four levels of cognitive input: (1) Passive (passive) In this state, the learner mainly receives information without any active processing. For example, when listening to a lecture or reading a textbook, the learner may only passively receive the content without deep thinking or interaction. (2) Active (active) The learner actively participates in the learning process at this stage, and processes information. This may include taking notes, participating in discussions or solving problems. Active learning helps deepen understanding of the material. (3) Constructive (constructive) At the constructive stage, the learner not only actively participates, but also integrates new knowledge with existing knowledge to construct their own understanding. This may involve creatively applying knowledge, analyzing and synthesizing to form new perspectives or models. (4) Interactive (interactive) The highest level of cognitive input, the learner interacts with others to jointly construct knowledge. Through group discussions, collaborative learning, and other ways, learners deepen their understanding through communication and challenge each other's perspectives.

[0121] In the embodiment of the present application, through the ICAP cognitive input framework, the cognitive input of the online discussion text is divided into four levels from low to high (passive (1)--active (2)--constructive (3)--interactive (4)).

[0122] S5-2. The correlation analysis tool is used to correlate the online discussion text, and the correlation between the online discussion text and the semantic complexity and cognitive input level is obtained.

[0123] According to the results of the cognitive input degree and the semantic complexity of the online discussion text obtained in step S4, the results are described in association using an association analysis tool, and the semantic complexity features of the online discussion text at different cognitive input levels based on psychological language attributes are obtained. The association analysis tool used in the embodiment of the present application is SPSS Modeler. SPSS Modeler is a tool for predictive analysis, data mining and text analysis. It allows users to create complex data streams through an intuitive graphical interface, and to perform data preprocessing, modeling, evaluation and deployment.

[0124] After obtaining the semantic complexity features of the online discussion text at different cognitive input levels based on psychological language attributes, the embodiment of the present application draws an error bar graph of the cognitive input level (X-axis) and the semantic complexity (Y-axis) to analyze the association between the online discussion text at different cognitive input levels and the semantic complexity. As shown in the error bar graph, Figure 6 It can be seen that as the cognitive input increases, the semantic complexity also increases as a whole. The highest cognitive input level is interactive, and in addition to considering the semantic complexity, the interaction in the discussion, such as the number of likes and comments, also needs to be considered.

[0125] In some embodiments, ANOVA (analysis of variance) tool can also be used to test the association description, and the test results are shown in Table 2:

[0126] Table 2 ANOVA test results

[0127]

[0128]

[0129] In step S5, the trained model automatically identifies the cognitive input level of the online discussion text, which specifically includes the following steps:

[0130] S5-3. Determine the target machine learning model of the training;

[0131] S5-4. Construct a training data set with the semantic complexity as the input feature vector and the cognitive input level as the model output;

[0132] S5-5. Train the model using the training data set to obtain a machine learning model capable of automatically identifying the cognitive input level of the online discussion text.

[0133] As Figure 7As shown, the embodiment of the present application quantifies the psychological and linguistic attributes such as familiarity F and abstractness A after conceptualizing the online discussion text P, and uses the calculated semantic complexity and comment like number as the feature vector (X feature vector), and uses the cognitive input degree as the prediction result (Y prediction result), to construct a data set. The best_model function in the pycaret library is called to compare various machine learning models to obtain the best model for automatically identifying the cognitive input of online discussion text. (In the embodiment of the present application, it is a random forest model, the recognition accuracy is 0.8416, the precision prec is 0.8366, the recall rate Recall is 0.8416, and the F1 score is 0.8386).

[0134] The present application can further calculate the semantic complexity of online discussion text and improve the analysis accuracy and automatic evaluation level of online discussion text by constructing an educational concept-oriented psychological and linguistic attribute-based "familiarity-abstractness (F-A)" conceptual representation model. Compared with the prior art, the model can objectively and accurately define and describe educational concepts, avoiding the lack of pertinence caused by corpus dependence in traditional methods. By quantitatively calculating psychological and linguistic attributes, the present application not only realizes in-depth understanding of the cognitive input state of learners, but also effectively improves the calculation accuracy of semantic complexity and the effectiveness and interpretability of automatic evaluation of cognitive input. Thus, a more reliable tool is provided for research and application in the field of education.

[0135] In addition, the multi-stage processing flow of the present application combines the advanced BERT model to ensure the semantic relevance of educational concepts and text, making subsequent analysis more precise. Based on the "familiarity-abstractness (F-A)" conceptual representation model of psychological and linguistic attributes and the cognitive input degree of online discussion text represented by the ICAP framework, the semantic complexity of online discussion text is calculated and the model is trained to evaluate its cognitive input degree. This method not only improves the interpretability and reduces the "black box" effect, but also provides a clear quantitative standard for the cognitive processing of educational concepts, which helps to better realize the automatic evaluation and analysis of online discussion text.

[0136] The embodiment also provides a computer program product which, when running on a computer, causes the computer to execute the above-mentioned related steps to realize the text semantic calculation and cognitive input evaluation method based on conceptual representation provided by the above-mentioned embodiment.

[0137] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the embodiments of the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions, and provide corresponding operation portal for user to choose authorization or refusal.

[0138] Those skilled in the art can understand that the modules in the device in the embodiments of the present application can be adaptively changed and arranged in one or more devices different from the embodiments. The modules or units or components in the embodiments of the present application can be combined into one module or unit or component, and furthermore can be divided into multiple sub-modules or sub-units or sub-components. All features disclosed in the specification (including the corresponding claims, abstract and drawings) and all processes or units of any method or device disclosed in this way can be combined in any combination, except that at least some of such features and / or processes or units are mutually exclusive. Unless explicitly stated otherwise, each feature disclosed in the specification (including the corresponding claims, abstract and drawings) can be replaced by an alternative feature providing the same, equivalent or similar purpose.

[0139] Through the description of the above embodiments, those skilled in the art can understand that, for the convenience and brevity of description, only the division of the above functional modules is taken as an example, and in actual application, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above.

[0140] Note that the logic and / or steps represented and / or otherwise described herein in flow chart form represent executable instructions, for example, that can be sequenced to implement logic functionality, and can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from a medium and execute the instructions. For purposes of this specification, a "computer-readable medium" can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. The computer-readable medium can specifically include the following: an electronic connection (an electronic device having one or more wires), a portable computer diskette (a magnetic device), a random-access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium upon which the program is printed, as the program can be electronically captured, for example, via an optical scanner, then compiled, interpreted, or otherwise processed in the electronic manner into a useable form, and then stored in the computer memory.

[0141] In addition, each of the various embodiments described in this specification has been described as if specific implementations only are envisioned. However, the person skilled in the art will understand that the various embodiments described in this specification can be combined with each other, unless specifically noted. In addition, the various embodiments described in this specification can be implemented in software and / or hardware. In addition, the various embodiments described in this specification can be implemented in a distributed system having one or more nodes that can be located in one or more locations. In addition, the various embodiments described in this specification can be implemented in a system that includes one or more devices that can be located in one or more locations. In addition, the various embodiments described in this specification can be implemented in a system that includes one or more devices that can be located in one or more locations.

[0142] It should be understood that aspects of the application can be implemented in hardware, software, firmware or a combination thereof. In the above embodiments, various steps or methods can be implemented in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and / or the like.

[0143] In the description of the present application, the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like, are intended to mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. The appearances of the above terms in various places in the specification are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0144] In addition, the terms "first", "second", and the like, in the description of the embodiments of the present application are used for descriptive purposes and not necessarily for describing a relative importance or a relative order of the technical features indicated in the embodiments. Therefore, the technical features defined with the terms "first", "second" and the like in the embodiments of the present application can explicitly or implicitly indicate that the embodiments include at least one of the technical features. In the description of the present application, the term "a plurality of" means at least two or two or more, for example, two, three, four, and the like, unless otherwise specifically limited in the embodiments.

[0145] In the embodiments of the present application, the terms "comprise", "contain" or any other variant thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed, or further includes elements inherent to such process, method, article or device. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of another identical element in the process, method, article or device including the element, in addition, the components, features, elements with the same name in different embodiments of the present application can have the same meaning or different meaning, and the specific meaning thereof should be determined in the interpretation of the specific embodiment or further combined with the context in the specific embodiment.

[0146] Although the embodiments of the present application have been shown and described above, it is understood that the above-described embodiments are exemplary and are not to be construed as limiting the present application, and that variations, modifications, substitutions and changes can be suggested to one skilled in the art without departing from the scope of the present application. Other embodiments of the present application will be apparent to those skilled in the art from consideration of the specification and practice of the present application. It is intended that the present application cover any and all variations of the application that come within the scope of the present application, along with all of the equivalents thereof. It is intended that the specification and examples be considered exemplary only, with the true scope and spirit of the application indicated by the following claims.

Claims

1. A method for text semantic computation and cognitive engagement assessment based on conceptual representation, characterized in that, The method comprises the following steps: constructing an education concept dictionary, matching education concepts in the online discussion text using the education concept dictionary to obtain candidate education concepts of the online discussion text; calculating semantic relevance of the online discussion text and the candidate education concepts using a BERT model, and filtering candidate education concepts with a semantic relevance lower than a preset relevance threshold; constructing an education concept familiarity-abstractness F-A model according to the education concept dictionary, and obtaining complexity and abstractness of the candidate education concepts from the education concept familiarity-abstractness F-A model; calculating semantic complexity of the online discussion text according to the complexity and abstractness of the candidate education concepts; performing correlation analysis on the cognitive input level and the semantic complexity of the online discussion text, and training a model to automatically identify the cognitive input level of the online discussion text; wherein the constructing of the education concept familiarity-abstractness F-A model according to the education concept dictionary and the obtaining of the complexity and abstractness of the candidate education concepts from the education concept familiarity-abstractness F-A model comprise the following steps: performing feature vectorization representation on the education concepts in the education concept dictionary to construct language feature vectors of the education concepts; performing regression analysis on the language feature vectors of the education concepts to determine the familiarity and abstractness of each education concept, and establishing an education concept familiarity-abstractness F-A model; obtaining the complexity and abstractness of the candidate education concepts in the education concept familiarity-abstractness F-A model.

2. The method of claim 1, wherein, The constructing of the education concept dictionary comprises the following steps: extracting a plurality of education concepts from a preset corpus; storing and organizing the education concepts using a Trie tree data structure to form an education concept dictionary.

3. The method of claim 1, wherein, The matching of the education concepts in the online discussion text using the education concept dictionary to obtain the candidate education concepts of the online discussion text comprises the following steps: performing word segmentation processing on the online discussion text; calculating the edit distance between the segmented online discussion text and each education concept in the education concept dictionary, and taking education concepts with an edit distance lower than a preset distance threshold as candidate education concepts of the online discussion text.

4. The method of claim 1, wherein, The calculating of the semantic relevance of the online discussion text and the candidate education concepts using the BERT model comprises the following steps: constructing a vector space of the online discussion text and the candidate education concepts using the BERT model; calculating the semantic relevance of each candidate education concept and the online discussion text using the following formula: ； wherein V P represents a vector space of online discussion texts, V Ci represents a vector space of each candidate educational concept, , j is the total number of candidate educational concepts.

5. The method of claim 1, wherein, The filtering of the candidate education concepts with a semantic relevance lower than a preset relevance threshold comprises the following steps: constructing a candidate education concept set according to the candidate education concepts; determining a preset relevance threshold according to the mean and standard deviation of the candidate education concept set.

6. The method of claim 1, wherein, The calculating of the semantic complexity of the online discussion text according to the complexity and abstractness of the candidate education concepts comprises the following steps: representing the online discussion text as a set of candidate education concepts; calculating the semantic complexity of the online discussion text using the following formula: ； where P represents the online discussion text, C i represents the i-th candidate educational concept in P, N represents the total number of candidate educational concepts, and i represents each candidate educational concept, ; W F represents the familiarity weight of the candidate educational concept, F ( C i ) represents the familiarity of the candidate educational concept in the educational concept familiarity-abstractness F-A model; W A represents the abstractness weight of the candidate educational concept, A ( C i ) represents the abstractness of the candidate educational concept in the educational concept familiarity-abstractness F-A model.

7. The method of claim 6, wherein the method further comprises: The familiarity weight and the abstractness weight of the candidate educational concept are calculated by a quasi-Newton method.

8. The method of claim 1, wherein, The correlation analysis of the cognitive input level and the semantic complexity of the online discussion text specifically includes the following steps: Cognitive input level coding is performed on the online discussion text. Correlation analysis tools are used to describe the correlation of the online discussion text, so as to obtain the correlation of the online discussion text with the semantic complexity and the cognitive input level.

9. The method of claim 1, wherein, The training model automatically identifies the cognitive input level of the online discussion text, specifically including the following steps A target machine learning model is determined for training; A training data set is constructed with the semantic complexity as an input feature vector and the cognitive input level as a model output; The model is trained using the training data set, so as to obtain a machine learning model capable of automatically identifying the cognitive input level of the online discussion text.

Images